A data center load elasticity adjustment potential evaluation method and device

Abstract

The application belongs to the field of data management, and discloses a data center load elasticity regulation potential evaluation method and device, which comprises the following steps: performing scale unification processing on operation information collected from a data center, combining an attention mechanism driven feature selection algorithm to screen out key indicators; designing and calculating the objective weights of each core evaluation indicator, combining a multi-dimensional coupling operation of an intuitionistic fuzzy comprehensive evaluation model to obtain the migration level of each task; calculating the proportion of the total power load of tasks determined as high and medium migration levels in the current total load of the data center, and when the proportion exceeds a preset threshold, using a baseline load curve of the data center to solve a multi-constraint optimization model and calculate the elasticity regulation interval of the power load; taking the generated evaluation data as independent evidence to perform data fusion; and according to a deviation correction type GBRT algorithm, correcting the fused evaluation results to output the load elasticity regulation potential evaluation results.

Description

Technical Field

[0001] This invention belongs to the field of data management, and in particular relates to a method and apparatus for assessing the load elasticity adjustment potential of a data center. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence technologies, the scale and energy consumption of data centers continue to grow. As the core of information infrastructure, the energy consumption of data centers has become a major global concern. Statistics show that data center energy consumption accounts for more than 2% of global total energy consumption and is still growing rapidly. Therefore, how to effectively reduce data center energy consumption and improve energy efficiency has become a critical issue that the industry urgently needs to address.

[0003] In related technologies, data center load regulation mainly relies on traditional resource scheduling strategies, such as virtual machine migration based on load balancing and load adjustment based on prediction. However, these methods have the following shortcomings: First, existing methods often focus only on single-dimensional resource optimization, lacking a comprehensive assessment of task portability, leading to inaccurate assessment of regulation potential; second, traditional methods do not fully consider the multi-dimensional characteristics of tasks (such as timeliness, data dependency, resource adaptability, etc.) during the evaluation process, making it difficult to accurately identify portable tasks; third, existing technologies often use simple linear models to calculate load regulation intervals, which cannot effectively handle complex nonlinear relationships and multi-constraint optimization problems; finally, existing methods have shortcomings in data fusion and evaluation accuracy optimization, resulting in low reliability of evaluation results. Summary of the Invention

[0004] In view of this, the present invention discloses a method and apparatus for assessing the load elasticity adjustment potential of data centers, which can solve the shortcomings of related technologies.

[0005] To achieve the above objectives, the present invention discloses the following technical solution: According to a first aspect of the present invention, a method for assessing the load resilience potential of a data center is proposed, comprising: An improved adaptive Z-score normalization method is used to perform scale-uniform processing on the operational information collected from the data center, and an attention-driven feature selection algorithm is combined to screen out key indicators to complete the preprocessing of the evaluation basis data; wherein, the operational information includes at least task characteristics, power load and cross-node resource configuration data; Based on the preprocessed data, three core evaluation indicators are designed: task timeliness, data dependence, and resource adaptability. The objective weights of each core evaluation indicator are calculated using the entropy weight method, and multi-dimensional coupling operations are performed using an intuitionistic fuzzy comprehensive evaluation model to obtain the transferability level of each task. The transferability level includes three levels: high, medium, and low. The total power load of tasks classified as high and medium migration level is calculated as a proportion of the current total load of the data center. When the proportion exceeds a preset threshold, the Attention-BiGRU model is used to predict the baseline load curve of the data center. The BO-NSGA-III algorithm is used to solve the multi-constraint optimization model with the goal of minimizing load fluctuations and maximizing migration benefits, and the elastic adjustment range of power load is calculated. The generated evaluation data is treated as an independent body of evidence, and data fusion is performed based on the DS evidence theory. The objective weights of each evidence body are calculated using the CRITIC method and consistency verification is performed. The fused evaluation results are then corrected using the bias-corrected GBRT algorithm to output the load resilience adjustment potential evaluation results.

[0006] According to a second aspect of the present invention, a data center load resilience assessment device is provided, the device comprising: Processing Unit: An improved adaptive Z-score normalization method is used to perform scale-unified processing on the operational information collected from the data center, and an attention-driven feature selection algorithm is used to filter out key indicators to complete the preprocessing of the evaluation base data; wherein, the operational information includes at least task characteristics, power load and cross-node resource configuration data; Design Unit: Based on the preprocessed data, three core evaluation indicators are designed: task timeliness, data dependency, and resource adaptability. The objective weights of each core evaluation indicator are calculated using the entropy weight method, and multi-dimensional coupling operations are performed using an intuitionistic fuzzy comprehensive evaluation model to obtain the transferability level of each task. The transferability level includes three levels: high, medium, and low. Calculation Unit: Calculates the proportion of the total power load of tasks classified as high and medium migration levels to the current total load of the data center. When the proportion exceeds a preset threshold, it uses the Attention-BiGRU model to predict the baseline load curve of the data center. Based on the BO-NSGA-III algorithm, it solves a multi-constraint optimization model with the objectives of minimizing load fluctuations and maximizing migration benefits, and calculates the elastic adjustment range of the power load. Fusion Unit: The generated evaluation data is treated as an independent body of evidence, and data fusion is performed based on DS evidence theory; Correction Unit: The objective weights of each evidence body are calculated using the CRITIC method and consistency verification is performed. The fused evaluation results are then corrected using the bias-corrected GBRT algorithm to output the load resilience adjustment potential evaluation results.

[0007] According to a third aspect of the present invention, an electronic device is provided, comprising: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in the first aspect by running the executable instructions.

[0008] According to a fourth aspect of the invention, a computer-readable storage medium is provided having computer instructions stored thereon that, when executed by a processor, implement the steps of the method as described in the first aspect.

[0009] As can be seen from the above technical solutions, the data center load elasticity adjustment potential assessment method disclosed in this invention is as follows: On the one hand, by adopting improved adaptive Z-score normalization to achieve data scale uniformity and combining it with an attention-driven feature selection algorithm to screen key indicators, redundant data can be effectively eliminated and outlier interference reduced. This solves the problems of coarse data preprocessing and low core feature identification in existing technologies, providing high-quality and highly targeted basic data for subsequent evaluation. On the other hand, the task transferability assessment is objective and comprehensive: This invention designs three core indicators: task timeliness, data dependence, and resource adaptability. It quantifies objective weights using the entropy weight method and combines them with an intuitionistic fuzzy comprehensive evaluation model to conduct multi-dimensional coupled calculations. This avoids the shortcomings of single task evaluation indicators and subjective weight setting in existing technologies, accurately determining the transferability level of each task, and providing a reliable decision-making basis for focusing on high-value adjustment tasks. Furthermore, the Attention-BiGRU model is used to accurately predict the baseline load, and the BO-NSGA-Ⅲ algorithm is combined to construct a multi-constraint model to calculate the adjustment range. At the same time, the DS evidence theory and the bias-corrected GBRT algorithm are used to optimize the evaluation accuracy. This can solve the problems of fuzzy calculation of load adjustment range and insufficient accuracy of evaluation results in the existing technology, realize the quantitative and accurate evaluation of load elasticity adjustment potential, and adapt to the dual practical needs of power grid dispatch and data center energy consumption optimization. Attached Figure Description

[0010] Figure 1 This is a flowchart illustrating an exemplary embodiment of a method for assessing the resilience and adaptability of data center loads. Figure 2 This is a schematic structural diagram of a device provided in an exemplary embodiment; Figure 3 This is a block diagram of a data center load resilience assessment device provided in an exemplary embodiment. Detailed Implementation

[0011] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of the present invention as detailed in the appended claims.

[0012] It should be noted that in other embodiments, the corresponding methods are not necessarily performed in the order shown and described in this invention. The method comprises steps. In some other embodiments, the method may include more or fewer steps than those described in this invention. Furthermore, a single step described in this invention may be broken down into multiple steps in other embodiments; and multiple steps described in this invention may be combined into a single step in other embodiments.

[0013] In today's digital age, data centers, as the core infrastructure of information technology, bear the burden of massive data storage, processing, and analysis tasks. With the explosive growth of data volume and the increasing complexity of business needs, the energy consumption problem of data centers has become increasingly prominent. In order to achieve efficient and energy-saving operation of data centers, load elasticity adjustment has become a key means, and accurate assessment of the load elasticity adjustment potential is the foundation and prerequisite for achieving effective adjustment.

[0014] In the current power load management of data centers, the following challenges are mainly faced: (1) Insufficient business correlation mining of task characteristics and load elasticity: Traditional assessment methods often process power load data in isolation, without establishing a dynamic correlation mechanism between task characteristics and load elasticity in combination with actual business operations, ignoring the direct impact of core business dimensions such as task operation status, resource consumption patterns, and cross-node migration requirements on load distribution. This results in load elasticity assessment remaining only at the static data statistics level, unable to adapt to the load adjustment needs under dynamic business changes, thus causing business problems such as insufficient targeting of adjustment strategies, redundant or insufficient resource allocation, and energy consumption control not meeting expectations. (2) Lack of a scientific system for judging task portability: Existing technologies have not established a multi-dimensional coupled quantitative standard for task portability, and only divide migration levels through a single business indicator or simple rules, failing to fully consider the interaction of core characteristics such as task timeliness, data dependence, and resource adaptability, making it difficult to accurately identify high-value portable tasks, resulting in insufficient mining of load adjustment potential, low resource allocation efficiency, and inability to meet the dual needs of efficient business operation and cost control. (3) Insufficient accuracy and reliability of load assessment: The baseline load forecast does not take into account the task migration characteristics and multi-source data correlation patterns in the business scenario. Traditional algorithms are poorly adaptable to load changes caused by business fluctuations, and the forecast error is large. At the same time, the business information value and consistency of each data source are not effectively quantified during the multi-source data fusion process, and there is a lack of reliable verification mechanism, which leads to significant deviations in the fusion results. This affects the accuracy of the load elasticity adjustment range calculation and cannot provide strong support for the stable operation of data center business, energy saving and consumption reduction and cost optimization.

[0015] To address the shortcomings of related technologies, this invention proposes a method for assessing the resilience and adaptability of data center loads.

[0016] Figure 1 This is a flowchart illustrating an exemplary embodiment of a method for assessing the load resilience potential of a data center. (Example:) Figure 1 As shown, this method, applied to a direct-drive permanent magnet synchronous generator, may include the following steps: Step 101: The improved adaptive Z-score normalization method is used to perform scale unification processing on the operation information collected from the data center, and the key indicators are screened out by the attention mechanism-driven feature selection algorithm to complete the preprocessing of the evaluation basis data; wherein, the operation information includes at least task characteristics, power load and cross-node resource configuration data.

[0017] Specifically, step 101 may include: Step 1011: Collect three key data types: data center task characteristics, power load, and cross-node resource configuration. The feature vector of each task can be represented as: Electricity load data is obtained through... Each hardware device node obtains the device. At any moment Power consumption , forming a matrix Cross-node resource configuration information for nodes The resource allocation vector is .

[0018] Step 1012: Improved adaptive Z-score normalization is used to eliminate data scale differences, and an attention mechanism is combined to select key indicators to complete preprocessing. First, the data distribution is analyzed to identify outliers, then the data is standardized, and finally, features are refined to remove redundant information, improving data quality and the relevance of the evaluation.

[0019] In one embodiment, the improved adaptive Z-score normalization method is used to perform scale unification processing on the operational information collected from the data center, including: analyzing the data distribution to identify outliers; performing standardization transformation on normal data points; and correcting the identified outliers according to a reasonable boundary range set by the data distribution, so that all data are normalized to a uniform scale.

[0020] For normal data, use the improved adaptive Z-score normalization formula. Perform standardization transformation. For outliers, set reasonable boundary ranges based on data distribution. in, , This is an empirical coefficient; outliers outside the range are corrected to boundary values.

[0021] Construct a feature selection model and normalize the task feature data matrix. Electricity load data matrix and cross-node resource configuration information data matrix Input model. The association weights between features and the evaluation target are calculated using a self-attention mechanism; for example, an attention weight matrix is ​​obtained from the task feature data. ,in Indicates the first The first task The weights of each feature are assigned. Based on the weights, non-decision features are removed, and core features such as task priority and equipment power consumption are retained, forming a compact evaluation dataset.

[0022] Traditional Z-score methods are sensitive to outliers, which can easily lead to data distortion after standardization. The improved method of this invention avoids the negative impact of outliers on the overall data distribution by identifying and correcting them, ensuring that all subsequent algorithm models are built on a data foundation with uniform scale and reliable quality, thereby improving the accuracy and stability of the final evaluation results.

[0023] Step 102: Based on the preprocessed data, design three core evaluation indicators: task timeliness, data dependence, and resource adaptability; use the entropy weight method to calculate the objective weight of each core evaluation indicator, and combine it with the intuitionistic fuzzy comprehensive evaluation model to perform multi-dimensional coupling calculations to obtain the transferability level of each task, which includes three levels: high, medium, and low.

[0024] Specifically, step 102 may include: Step 1021: Design three quantifiable evaluation indicators for data center tasks: timeliness, data dependency, and resource adaptability. Define evaluation criteria for each indicator to ensure that the indicators are aligned with actual operations and to quantify the feasibility of data center migration.

[0025] In one embodiment, the task timeliness index is quantified by the relationship between the task deadline, current progress and expected migration time; the data dependency index is quantified by the frequency and degree of interaction between the task and external nodes; and the resource adaptability index is quantified by the degree of matching between the remaining resources of the target node in four types of resources (CPU, memory, storage and network) and the task requirements.

[0026] First, focusing on the redundancy of the task migration time window, a timeliness indicator is designed, and the task is set... The deadline is The current progress corresponds to the time period. The estimated migration time is Then the timeliness index For example, if task A's For 10 hours, It lasts for 3 hours. If it is 2 hours, then The higher the indicator value, the lower the risk of migration failure due to insufficient time.

[0027] Secondly, statistical tasks Frequency of interaction with external nodes and degree of correlation Data dependency metrics .

[0028]

[0029] in, , These are weighting coefficients set according to the actual situation.

[0030] Assuming task B times / hour , , ,but The smaller the indicator value, the weaker the correlation between the task and external data, the lower the data synchronization cost and the higher the efficiency after migration, and the stronger the migration feasibility.

[0031] Next, set the task. The requirements for CPU, memory, storage, and network are respectively , , , Target node The corresponding remaining resources are respectively , , , The resource adaptability index is: .

[0032]

[0033] in, These are the weighting coefficients for each resource.

[0034] For example, task C requires the following CPU, memory, storage, and network resources respectively: , , , The remaining resources corresponding to the target node are as follows: , , , , ,but The higher the indicator value, the better the target node can meet the task operation requirements, and the higher the stability after migration.

[0035] In this embodiment, by quantifying from three key dimensions—timeliness, data dependence, and resource adaptability—the limitations of existing technologies that rely solely on single or subjective experience are overcome. This allows the assessment of task migration feasibility to move from qualitative to quantitative, and from vague to precise, laying a solid foundation for accurately selecting high-potential tasks with practical adjustment value.

[0036] Step 1022: The objective weights of the three types of indicators are calculated using the entropy weight method. The quantification is completed through three steps: indicator data standardization, information entropy calculation, and weight solution, so that the weight allocation matches the actual distinguishing ability of the indicators.

[0037] In one embodiment, the step of calculating the objective weights of each of the core evaluation indicators using the entropy weight method includes: constructing an initial decision matrix according to the task and indicator dimensions; standardizing the positive and negative indicators in the initial decision matrix to form a standardized matrix; calculating the information entropy and difference coefficient of each indicator in the standardized matrix; and determining the final objective weight of each indicator based on the difference coefficient.

[0038] Construct a decision matrix based on the dimensions of "tasks and indicators". ,in , , .

[0039] For positive indicators The standardized formula is used: ; For negative indicators The formula used is: Forming a standardized matrix .

[0040] Based on the normalized matrix Calculate the information entropy of each indicator. The formula for information entropy is: ; in, , .

[0041] The smaller the information entropy value, the greater the difference in indicator data, the more significant the role in distinguishing task migration levels, and the higher the contribution in the comprehensive evaluation.

[0042] Step 2.2.3: Calculate the index difference coefficient using information entropy. The final weights of each indicator are calculated based on the proportion of the difference coefficient. , weight value Records are associated with the corresponding indicators to form an indicator weight vector. .

[0043] In this embodiment, the entropy weight method automatically calculates weights based on the degree of variation of each indicator's data, avoiding the subjective arbitrariness of manual weighting. This allows indicators with stronger discriminative capabilities and richer data information to be assigned higher weights in the comprehensive evaluation, thus making the final task level determination result more scientific and reliable.

[0044] Step 1023: Combine the indicator weight vector and fuzzy number to conduct a comprehensive evaluation. By determining the fuzzy relationship between the indicator and the migration level, the fuzzy synthesis operator is used to perform coupling operations, and the task level is determined based on the principle of maximum membership.

[0045] In one embodiment, the step of obtaining the transferability level of each task by performing multi-dimensional coupling calculations using an intuitionistic fuzzy comprehensive evaluation model includes: generating an intuitionistic fuzzy relation matrix reflecting the membership degree of each indicator value corresponding to different transferability levels based on historical data mining; using an intuitionistic fuzzy weighted average operator to fuse the objective weights with the intuitionistic fuzzy relation matrix to obtain a comprehensive evaluation vector for each task corresponding to each transferability level; and determining the final transferability level of the task from the comprehensive evaluation vector according to the principle of maximum membership degree.

[0046] Extract the core indicator values ​​and final migration level results of historical tasks, statistically analyze the frequency of each indicator value corresponding to the "high, medium and low" levels, convert the frequency into membership degree, and automatically generate intuitive fuzzy numbers.

[0047] For example, for a certain indicator, the frequency of its occurrence at higher levels is statistically analyzed. Then it is converted into membership degree. for This process is repeated for other levels of membership, resulting in a standardized "indicator-level" relationship matrix. .

[0048] The intuitionistic fuzzy weighted average synthesis operator is used to transform the index weight vector obtained by the entropy weight method. Fuzzy Relationship Matrix Generated by Data Mining Perform fusion calculation. Let the vector of the comprehensive intuitive fuzzy evaluation result be... ,in ( (Indicates high, medium, and low migration levels).

[0049] The task migration level is determined based on the principle of maximum membership, and the comprehensive membership value of the task at each level is compared. .like If the value is at its maximum, the task is considered to have a high migrationability level, meaning that migration offers significant potential benefits; if... At its highest level, it is a medium-transferable level, and cost control is necessary; if The highest level is considered low-migration, indicating a high migration risk and therefore not recommended for implementation.

[0050] In this embodiment, by combining historical data to generate an intuitionistic fuzzy relation matrix, the model can characterize the complex nonlinear relationship between indicators and levels. Furthermore, through fuzzy synthesis operations, it can comprehensively consider all indicator information, and even when some indicators exhibit fluctuations or noise, it can still yield robust comprehensive evaluation results, making the judgment logic closer to the decision-making process of human experts.

[0051] Step 103: Calculate the proportion of the total power load of tasks classified as high and medium migration levels to the current total load of the data center. When the proportion exceeds a preset threshold, use the Attention-BiGRU model to predict the baseline load curve of the data center. Solve the multi-constraint optimization model with the goal of minimizing load fluctuations and maximizing migration benefits based on the BO-NSGA-III algorithm, and calculate the elastic adjustment range of the power load.

[0052] Specifically, step 103 may include: Step 1031: Taking high- and medium-level transferable tasks as the core adjustment objects, filter the task list, extract related load data, calculate the load ratio to determine the adjustment space, locate tasks with transfer value, and quantify the occupied load resources.

[0053] In one embodiment, the calculation of the proportion of the total power load of tasks classified as high and medium migration level to the current total load of the data center includes: extracting a list of high and medium migration level tasks and associating them with the hardware device information of the executing tasks; extracting the real-time power of the hardware devices from the power load data, summarizing them by task to obtain the total power load; and using the ratio of the total power load to the current total load of the data center as the proportion.

[0054] Based on the level determination result in step 102, extract the high- and medium-level task lists, and associate them with unique task identifiers, execution nodes, and hardware device information. Let the high- and medium-level task sets be... For the task Record relevant information. Simultaneously remove tasks marked "uninterruptible" due to special business needs, forming a set of tasks with practical migration feasibility. .

[0055] Based on the equipment association information for high- and medium-level tasks, real-time power consumption data of the corresponding hardware devices is extracted from the power load data. For tasks... Summarize the workload of each individual task. ,in For the task The collection time points of relevant equipment. The total power load of this type of task is obtained by summing the loads of all target tasks. Synchronously record the current total load of the data center .

[0056] Calculate the proportion of high- and medium-level task loads in the total data center load: .

[0057] like , To determine the appropriate threshold, initiate the baseline load forecasting process; if The output adjustment space is insufficient, and it is recommended to prioritize optimizing inefficient equipment.

[0058] In this embodiment, by calculating the load percentage of high- and medium-level portable tasks, it is possible to intuitively determine whether the current data center has sufficient load adjustment potential. Setting a threshold avoids initiating unnecessary complex calculations when adjustment space is insufficient, improving the efficiency and practicality of the evaluation process and enabling on-demand resource allocation.

[0059] Step 1032: Construct an Attention-BiGRU load prediction model, select multi-dimensional input features, build a "BiGRU + self-attention" network structure, train the model to output a baseline load curve, and accurately predict the baseline load of the data center when there is no task migration.

[0060] In one embodiment, the Attention-BiGRU model includes: an input layer for receiving a multi-dimensional feature vector containing historical load, environmental parameters, task execution intensity, and date type; a BiGRU layer for bidirectionally extracting time-dependent features from the multi-dimensional feature vector; a self-attention layer for assigning weights to features at different time steps to enhance key scene features; and a fully connected layer for mapping the weighted features to a baseline load prediction value.

[0061] The model input matrix is ​​constructed by selecting multi-dimensional features from the historical operational database, including historical load data. Environmental parameters Task execution intensity Date type Features are normalized and missing value imputation is performed, and they are organized into a model-recognizable input format according to the "feature, time" dimension to form feature vectors. , For a point in time.

[0062] A prediction model is constructed, consisting of an input layer, a BiGRU layer, a self-attention layer, and a fully connected layer. The input layer processes the feature vectors... The model is transformed into a dimension suitable for model processing; the BiGRU layer captures the load time dependency bidirectionally; the self-attention layer strengthens the weights of key scene features; and the fully connected layer maps the weighted features to load prediction values ​​to complete the model construction.

[0063] The training and test sets are divided using historical data, and the loss function is set as follows: The optimizer is By iteratively optimizing the model parameters After the model is trained, input the parameters of the current business scenario, and it will output the hourly baseline load curve for a future period of time. .

[0064] In this embodiment, the BiGRU layer can bidirectionally capture the temporal dependencies of load data; the self-attention mechanism can dynamically emphasize key historical information most relevant to the prediction time (such as specific business peaks), effectively reducing prediction errors caused by complex business fluctuations. This deep learning model, which combines spatiotemporal features, provides a more accurate estimate of baseline load compared to traditional time series forecasting methods, offering a reliable benchmark for subsequent adjustment interval calculations.

[0065] Step 1033 sets two objectives: "minimizing electricity load fluctuations" and "maximizing task migration benefits." The objective function for electricity load fluctuations is:

[0066] in The baseline load curve, This refers to the adjusted electrical load.

[0067] The objective function for task migration benefits is:

[0068] in For the task The electricity saved by relocation For the task Migration costs.

[0069] This paper proposes a method to solve a multi-objective optimization problem based on the NSGA-III algorithm, introducing Bayesian optimization to adaptively adjust key parameters such as crossover probability and mutation probability. The baseline load curve is then used. Task load data The optimization model is input with constraint parameters, and the BO-NSGA-Ⅲ algorithm is called to obtain the Pareto optimal solution set. The extreme values ​​of load regulation are extracted from the optimal solutions, and the flexible load regulation range is calculated by combining this with the baseline load curve. .

[0070] Step 104: Treat the generated evaluation data as independent evidence and perform data fusion based on DS evidence theory.

[0071] Assume that the data sources include the grade determination results, load assessment data, etc. , , , Assign credibility weights to each piece of evidence. , , , Data conflicts are handled using the D-S evidence synthesis rule, with the basic probability allocation function set as follows: , , , The synthesized basic probability assignment function is: This integrates scattered data into a unified comprehensive evaluation dataset.

[0072] Step 105: Calculate the objective weights of each evidence body using the CRITIC method and perform consistency verification. Then, correct the fused evaluation results according to the bias-corrected GBRT algorithm to output the load resilience adjustment potential evaluation results.

[0073] Construct an evaluation matrix using core indicators from various data sources. Calculate the standard deviation of the index ,in .

[0074] Calculate the amount of information in the data source using a formula. Determine objective weights based on the proportion of information content. .

[0075]

[0076] in, For data source and The correlation coefficient between them.

[0077] Construct a consistency verification model to compare the deviations between the merged data and the independent data sources. Let the deviation function be... The significance test is used to determine whether the deviation is reasonable. If the deviation is within a reasonable range and the proportion of data points meets the standard, and the abnormal data meets the standard after correction, the data is considered reliable; otherwise, the process is returned to the data source analysis stage.

[0078] A bias-corrected GBRT model is constructed using reliable fused data as input. Let the model's predicted values ​​be... The actual value is The mean squared error loss function is usually chosen: ; in, For the sample size, and They are the first The predicted and actual values ​​of a sample.

[0079] System biases are corrected through model training, and model parameters are optimized using gradient descent. In each iteration, the loss function is calculated with respect to the model parameters. gradient Then, update the parameters in the opposite direction of the gradient, using the following formula: , in, The learning rate controls the step size for each parameter update.

[0080] The above process is repeated until the loss function converges or the preset maximum number of iterations is reached. The trained bias-corrected GBRT model can provide more accurate task migration level determinations and power load predictions for new input data, thus providing more precise and reliable support for data center task migration decisions and power load management.

[0081] The data center load resilience assessment method disclosed in this invention, on the one hand, achieves data scale uniformity by employing improved adaptive Z-score normalization and combines it with an attention-driven feature selection algorithm to screen key indicators. This effectively eliminates redundant data and reduces outlier interference, solving the problems of coarse data preprocessing and low core feature identification in existing technologies, providing high-quality and highly targeted basic data for subsequent assessment. On the other hand, the task mobility assessment is objective and comprehensive: this invention designs three core indicators—task timeliness, data dependence, and resource adaptability—and quantifies objective weights using the entropy weight method and combines them with an intuitionistic fuzzy comprehensive evaluation model to conduct multi-dimensional coupled calculations. This avoids the shortcomings of single task assessment indicators and subjective weight setting in existing technologies, accurately determining the transferability level of each task, and providing a reliable decision-making basis for focusing on high-value adjustment tasks. Furthermore, the Attention-BiGRU model is used to accurately predict the baseline load, and the BO-NSGA-Ⅲ algorithm is combined to construct a multi-constraint model to calculate the adjustment range. At the same time, the DS evidence theory and the bias-corrected GBRT algorithm are used to optimize the evaluation accuracy. This can solve the problems of fuzzy calculation of load adjustment range and insufficient accuracy of evaluation results in the existing technology, realize the quantitative and accurate evaluation of load elasticity adjustment potential, and adapt to the dual practical needs of power grid dispatch and data center energy consumption optimization.

[0082] Figure 2 This is a schematic structural diagram of a device provided in an exemplary embodiment. Please refer to... Figure 2 At the hardware level, the device includes a processor 202, an internal bus 204, a network interface 206, memory 208, and non-volatile memory 210, and may also include other hardware required for its functions. One or more embodiments of the present invention can be implemented in software, for example, the processor 202 reads the corresponding computer program from the non-volatile memory 210 into memory 208 and then runs it. Of course, in addition to software implementation, one or more embodiments of the present invention do not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0083] Please refer to Figure 3A data center load resilience assessment device can be applied to, for example... Figure 3 The device shown, in order to implement the technical solution of the present invention, includes: The processing unit 301 is used to perform scale-unified processing on the operational information collected from the data center using an improved adaptive Z-score normalization method, and to filter out key indicators by combining an attention mechanism-driven feature selection algorithm to complete the preprocessing of the evaluation base data; wherein, the operational information includes at least task characteristics, power load and cross-node resource configuration data. Design unit 302 is used to design three core evaluation indicators based on preprocessed data: task timeliness, data dependency, and resource adaptability; calculate the objective weight of each core evaluation indicator using the entropy weight method; and perform multi-dimensional coupling calculations using an intuitionistic fuzzy comprehensive evaluation model to obtain the transferability level of each task, which includes three levels: high, medium, and low. The calculation unit 303 is used to calculate the proportion of the total power load of tasks judged as high and medium migration level to the current total load of the data center. When the proportion exceeds a preset threshold, the Attention-BiGRU model is used to predict the baseline load curve of the data center. The BO-NSGA-III algorithm is used to solve the multi-constraint optimization model with the goal of minimizing load fluctuation and maximizing migration benefits, and the elastic adjustment range of power load is calculated. The fusion unit 304 is used to treat the generated evaluation data as independent evidence and perform data fusion based on the DS evidence theory. The correction unit 305 is used to calculate the objective weights of each evidence body using the CRITIC method and perform consistency verification, and to correct the fused evaluation results according to the deviation-corrected GBRT algorithm, so as to output the load elasticity adjustment potential evaluation results.

[0084] Optionally, the processing unit 301 is specifically used for: Analyze data distribution to identify outliers; Perform standardization transformation on normal data points; For identified outliers, corrections are made based on reasonable boundary ranges set according to the data distribution, so that all data are normalized to a uniform scale.

[0085] Optionally, the task timeliness indicator is quantified by the relationship between the task deadline, current progress, and estimated migration time; the data dependency indicator is quantified by the frequency and degree of interaction between the task and external nodes; and the resource adaptability indicator is quantified by the degree of matching between the remaining resources of the target node in four types of resources (CPU, memory, storage, and network) and the task requirements.

[0086] Furthermore, the design unit 302 is specifically used for: Construct an initial decision matrix based on task and indicator dimensions; The positive and negative indices in the initial decision matrix are standardized to form a standardized matrix. Calculate the information entropy and difference coefficient of each indicator in the standardized matrix; The final objective weight of each indicator is determined based on the difference coefficient.

[0087] Furthermore, the design unit 302 is specifically used for: Based on historical data mining, an intuitive fuzzy relation matrix is ​​generated that reflects the membership degree of each indicator value corresponding to different transferable levels. An intuitionistic fuzzy weighted average operator is used to fuse the objective weights with the intuitionistic fuzzy relation matrix to obtain a comprehensive evaluation vector for each transferable level of the task. Based on the principle of maximum membership, the final transferability level of the task is determined from the comprehensive evaluation vector.

[0088] Optionally, the computing unit 303 is specifically used for: Extract a list of high and medium scalable tasks and associate it with the hardware device information on which they are executed; The real-time power of the hardware device is extracted from the power load data, and the total power load is obtained by summarizing the data by task. The ratio is the ratio of the total power load to the current total load of the data center.

[0089] Optionally, the Attention-BiGRU model includes: The input layer is used to receive multi-dimensional feature vectors containing historical load, environmental parameters, task execution intensity, and date type. The BiGRU layer is used to extract time-dependent features bidirectionally from the multi-dimensional feature vector; A self-attention layer is used to assign weights to features at different time steps, thereby enhancing key scene features. A fully connected layer is used to map weighted features to baseline load forecasts.

[0090] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, which can take the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email sending and receiving device, game console, tablet computer, wearable device, or any combination of these devices.

[0091] In a typical configuration, a computer includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0092] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0093] Computer-readable media, including both permanent and non-permanent, removable and non-removable media, can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage, quantum memory, graphene-based storage media or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0094] For any other form of computer-readable medium (or computer-readable storage medium) as described above, computer instructions may be stored thereon, which, when executed by a processor, implement one or more of the above embodiments, thereby realizing the technical solution of the present invention.

[0095] The present invention also proposes a computer program that, when executed by a processor, implements one or more of the embodiments described above, thereby realizing the technical solution of the present invention. This computer program may be specifically recorded on the above-described or other computer-readable media, and the present invention does not impose any limitations on this.

[0096] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0097] The foregoing has described specific embodiments of the invention. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired results. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] The terminology used in one or more embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in one or more embodiments of the invention and in the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0099] It should be understood that although the terms first, second, third, etc., may be used to describe various information in one or more embodiments of the present invention, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, first information may also be referred to as second information without departing from the scope of one or more embodiments of the present invention, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination."

[0100] The above description is merely a preferred embodiment of one or more embodiments of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of one or more embodiments of the present invention should be included within the protection scope of one or more embodiments of the present invention.

Claims

1. A method for assessing the load resilience potential of a data center, characterized in that, include: An improved adaptive Z-score normalization method is used to perform scale-uniform processing on the operational information collected from the data center, and an attention-driven feature selection algorithm is combined to screen out key indicators to complete the preprocessing of the evaluation basis data; wherein, the operational information includes at least task characteristics, power load and cross-node resource configuration data; Based on the preprocessed data, three core evaluation indicators are designed: task timeliness, data dependence, and resource adaptability. The objective weights of each core evaluation indicator are calculated using the entropy weight method, and multi-dimensional coupling operations are performed using an intuitionistic fuzzy comprehensive evaluation model to obtain the transferability level of each task. The transferability level includes three levels: high, medium, and low. The total power load of tasks classified as high and medium migration level is calculated as a proportion of the current total load of the data center. When the proportion exceeds a preset threshold, the Attention-BiGRU model is used to predict the baseline load curve of the data center. The BO-NSGA-III algorithm is used to solve the multi-constraint optimization model with the goal of minimizing load fluctuations and maximizing migration benefits, and the elastic adjustment range of power load is calculated. The generated evaluation data is treated as an independent body of evidence, and data fusion is performed based on the DS evidence theory. The objective weights of each evidence body are calculated using the CRITIC method and consistency verification is performed. The fused evaluation results are then corrected using the bias-corrected GBRT algorithm to output the load resilience adjustment potential evaluation results.

2. The method according to claim 1, characterized in that, The improved adaptive Z-score normalization method is used to perform scale-unified processing on the operational information collected from the data center, including: Analyze data distribution to identify outliers; Perform standardization transformation on normal data points; For identified outliers, corrections are made based on reasonable boundary ranges set according to the data distribution, so that all data are normalized to a uniform scale.

3. The method according to claim 1, characterized in that, The task timeliness index is quantified by the relationship between the task deadline, current progress, and estimated migration time; the data dependency index is quantified by the frequency and degree of interaction between the task and external nodes; and the resource adaptability index is quantified by the degree of matching between the remaining resources of the target node in four types of resources (CPU, memory, storage, and network) and the task requirements.

4. The method according to claim 3, characterized in that, The calculation of the objective weights of each core evaluation indicator using the entropy weight method includes: Construct an initial decision matrix based on task and indicator dimensions; The positive and negative indices in the initial decision matrix are standardized to form a standardized matrix. Calculate the information entropy and difference coefficient of each indicator in the standardized matrix; The final objective weight of each indicator is determined based on the difference coefficient.

5. The method according to claim 4, characterized in that, The method of combining the intuitionistic fuzzy comprehensive evaluation model with multi-dimensional coupling operations to obtain the transferability level of each task includes: Based on historical data mining, an intuitive fuzzy relation matrix is ​​generated that reflects the membership degree of each indicator value corresponding to different transferable levels. An intuitionistic fuzzy weighted average operator is used to fuse the objective weights with the intuitionistic fuzzy relation matrix to obtain a comprehensive evaluation vector for each transferable level of the task. Based on the principle of maximum membership, the final transferability level of the task is determined from the comprehensive evaluation vector.

6. The method according to claim 1, characterized in that, The calculation determines the proportion of the total power load of tasks classified as high and medium migration-level to the current total load of the data center, including: Extract a list of high and medium scalable tasks and associate it with the hardware device information on which they are executed; The real-time power of the hardware device is extracted from the power load data, and the total power load is obtained by summarizing the data by task. The ratio is the ratio of the total power load to the current total load of the data center.

7. The method according to claim 1, characterized in that, The Attention-BiGRU model includes: The input layer is used to receive multi-dimensional feature vectors containing historical load, environmental parameters, task execution intensity, and date type. The BiGRU layer is used to extract time-dependent features bidirectionally from the multi-dimensional feature vector; The self-attention layer is used to assign weights to features at different time steps, thereby enhancing key scene features. A fully connected layer is used to map weighted features to baseline load forecasts.

8. A device for assessing the load resilience potential of a data center, characterized in that, The device includes: Processing Unit: An improved adaptive Z-score normalization method is used to perform scale-unified processing on the operational information collected from the data center, and an attention-driven feature selection algorithm is used to filter out key indicators to complete the preprocessing of the evaluation base data; wherein, the operational information includes at least task characteristics, power load and cross-node resource configuration data; Design Unit: Based on the preprocessed data, three core evaluation indicators are designed: task timeliness, data dependency, and resource adaptability. The objective weights of each core evaluation indicator are calculated using the entropy weight method, and multi-dimensional coupling operations are performed using an intuitionistic fuzzy comprehensive evaluation model to obtain the transferability level of each task. The transferability level includes three levels: high, medium, and low. Calculation Unit: Calculates the proportion of the total power load of tasks classified as high and medium migration levels to the current total load of the data center. When the proportion exceeds a preset threshold, it uses the Attention-BiGRU model to predict the baseline load curve of the data center. Based on the BO-NSGA-III algorithm, it solves a multi-constraint optimization model with the objectives of minimizing load fluctuations and maximizing migration benefits, and calculates the elastic adjustment range of the power load. Fusion Unit: The generated evaluation data is treated as an independent body of evidence, and data fusion is performed based on the DS evidence theory; Correction Unit: The objective weights of each evidence body are calculated using the CRITIC method and consistency verification is performed. The fused evaluation results are then corrected using the bias-corrected GBRT algorithm to output the load resilience adjustment potential evaluation results.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor implements the steps of the method as described in any one of claims 1-7 by executing the executable instructions.

10. A computer-readable storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the method as described in any one of claims 1-7.

Images